![]() ![]() ![]() DeepMind used a tournament system with its modded version of Quake III Arena that pitted 40 human players against the company’s trained AI to calculate the AI’s skill levels. “In order to learn how to play tactically and collaborate with their teammates, these agents must rely on feedback from the game outcomes - without any teacher or coach showing them what to do.” “What makes these results so exciting is that these agents perceive their environment from a first-person perspective, just as a human player would,” DeepMind research scientist Thore Graepel said in a statement. This resulted in “decentralized control within a team” of AI agents, according to the paper. Instead, they learned independently from pixel data and a game score. Illustration of AI agents playing 'Quake III Arena.' Credit: DeepMindĬrucially, the AI agents in DeepMind’s Quake III Arena study had no access to game information that a human player wouldn’t, and didn’t learn from each other. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |